[Qemu-devel] [PATCH v4 0/3] Simpletrace v2: Support multiple args, strings.

[Qemu-devel] [PATCH v4 0/3] Simpletrace v2: Support multiple args, strings.

[Harsh Prateek Bora]

Existing simpletrace backend allows to trace at max 6 args and does not
support strings. This newer tracelog format gets rid of fixed size records
and therefore allows to trace variable number of args including strings.

Sample trace:
v9fs_version 0.000 tag=0xffff id=0x64 msize=0x2000 version=9P2000.L
v9fs_version_return 6.705 tag=0xffff id=0x64 msize=0x2000 version=9P2000.L
v9fs_attach 174.467 tag=0x1 id=0x68 fid=0x0 afid=0xffffffffffffffff
uname=nobody aname=
v9fs_attach_return 4720.454 tag=0x1 id=0x68 type=0xffffffffffffff80
version=0x4f2a4dd0  path=0x220ea6

v4:
- removed unused safe_strlen interface (missed in v3).

v3:
- Addressed Stefan's review comments on v2 series

v2:
- moved elimination of st_print_trace to 1/3 patch
- use sizeof(TraceRecord) for #define ST_REC_HDR_LEN

v1:
- Simpletrace v2 for the new (pythonized) tracetool

Harsh Prateek Bora (3):
  monitor: remove unused do_info_trace
  Simpletrace v2: Support multiple arguments, strings.
  Update simpletrace.py for new log format

 monitor.c                           |   16 ---
 scripts/simpletrace.py              |  116 ++++++++++------
 scripts/tracetool/backend/simple.py |   84 +++++++++---
 trace/simple.c                      |  249 ++++++++++++++++++++---------------
 trace/simple.h                      |   38 ++++--
 5 files changed, 312 insertions(+), 191 deletions(-)

-- 
1.7.10.2

[Qemu-devel] [PATCH v4 1/3] monitor: remove unused do_info_trace

[Harsh Prateek Bora]

Going forward with simpletrace v2 variable size trace records, we cannot
have a generic function to print trace event info and therefore this
interface becomes invalid.

As per Stefan Hajnoczi:

"This command is only available from the human monitor.  It's not very
useful because it historically hasn't been able to pretty-print events
or show them in the right order (we use a ringbuffer but it prints
them out from index 0).

Therefore, I don't think we're under any obligation to keep this
command around.  No one has complained about it's limitations - I
think this is a sign that no one has used it.  I'd be okay with a
patch that removes it."

Ref: http://lists.gnu.org/archive/html/qemu-devel/2012-01/msg01268.html

Signed-off-by: Harsh Prateek Bora <harsh@linux.vnet.ibm.com>
---
 monitor.c      |   16 ----------------
 trace/simple.c |   18 ------------------
 trace/simple.h |    1 -
 3 files changed, 35 deletions(-)

diff --git a/monitor.c b/monitor.c
index a3bc2c7..1fb00ab 100644
--- a/monitor.c
+++ b/monitor.c
@@ -795,13 +795,6 @@ static void do_info_cpu_stats(Monitor *mon)
 }
 #endif
 
-#if defined(CONFIG_TRACE_SIMPLE)
-static void do_info_trace(Monitor *mon)
-{
-    st_print_trace((FILE *)mon, &monitor_fprintf);
-}
-#endif
-
 static void do_trace_print_events(Monitor *mon)
 {
     trace_print_events((FILE *)mon, &monitor_fprintf);
@@ -2568,15 +2561,6 @@ static mon_cmd_t info_cmds[] = {
         .help       = "show roms",
         .mhandler.info = do_info_roms,
     },
-#if defined(CONFIG_TRACE_SIMPLE)
-    {
-        .name       = "trace",
-        .args_type  = "",
-        .params     = "",
-        .help       = "show current contents of trace buffer",
-        .mhandler.info = do_info_trace,
-    },
-#endif
     {
         .name       = "trace-events",
         .args_type  = "",
diff --git a/trace/simple.c b/trace/simple.c
index b4a3c6e..b64bcf4 100644
--- a/trace/simple.c
+++ b/trace/simple.c
@@ -291,24 +291,6 @@ void st_print_trace_file_status(FILE *stream, int (*stream_printf)(FILE *stream,
                   trace_file_name, trace_fp ? "on" : "off");
 }
 
-void st_print_trace(FILE *stream, int (*stream_printf)(FILE *stream, const char *fmt, ...))
-{
-    unsigned int i;
-
-    for (i = 0; i < TRACE_BUF_LEN; i++) {
-        TraceRecord record;
-
-        if (!get_trace_record(i, &record)) {
-            continue;
-        }
-        stream_printf(stream, "Event %" PRIu64 " : %" PRIx64 " %" PRIx64
-                      " %" PRIx64 " %" PRIx64 " %" PRIx64 " %" PRIx64 "\n",
-                      record.event, record.x1, record.x2,
-                      record.x3, record.x4, record.x5,
-                      record.x6);
-    }
-}
-
 void st_flush_trace_buffer(void)
 {
     flush_trace_file(true);
diff --git a/trace/simple.h b/trace/simple.h
index 466e75b..6b5358c 100644
--- a/trace/simple.h
+++ b/trace/simple.h
@@ -29,7 +29,6 @@ void trace3(TraceEventID event, uint64_t x1, uint64_t x2, uint64_t x3);
 void trace4(TraceEventID event, uint64_t x1, uint64_t x2, uint64_t x3, uint64_t x4);
 void trace5(TraceEventID event, uint64_t x1, uint64_t x2, uint64_t x3, uint64_t x4, uint64_t x5);
 void trace6(TraceEventID event, uint64_t x1, uint64_t x2, uint64_t x3, uint64_t x4, uint64_t x5, uint64_t x6);
-void st_print_trace(FILE *stream, fprintf_function stream_printf);
 void st_print_trace_file_status(FILE *stream, fprintf_function stream_printf);
 void st_set_trace_file_enabled(bool enable);
 bool st_set_trace_file(const char *file);
-- 
1.7.10.2

[Qemu-devel] [PATCH v4 2/3] Simpletrace v2: Support multiple arguments, strings.

[Harsh Prateek Bora]

Existing simpletrace backend allows to trace at max 6 args and does not
support strings. This newer tracelog format gets rid of fixed size records
and therefore allows to trace variable number of args including strings.

Sample trace with strings:
v9fs_version 0.000 tag=0xffff id=0x64 msize=0x2000 version=9P2000.L
v9fs_version_return 6.705 tag=0xffff id=0x64 msize=0x2000 version=9P2000.L

Signed-off-by: Harsh Prateek Bora <harsh@linux.vnet.ibm.com>
---
 scripts/tracetool/backend/simple.py |   84 ++++++++++---
 trace/simple.c                      |  231 +++++++++++++++++++++--------------
 trace/simple.h                      |   37 ++++--
 3 files changed, 237 insertions(+), 115 deletions(-)

diff --git a/scripts/tracetool/backend/simple.py b/scripts/tracetool/backend/simple.py
index fbb5717..24e02ac 100644
--- a/scripts/tracetool/backend/simple.py
+++ b/scripts/tracetool/backend/simple.py
@@ -15,9 +15,16 @@ __email__      = "stefanha@linux.vnet.ibm.com"
 
 from tracetool import out
 
+def is_string(arg):
+    strtype = ('const char*', 'char*', 'const char *', 'char *')
+    if arg.lstrip().startswith(strtype):
+        return True
+    else:
+        return False
 
 def c(events):
     out('#include "trace.h"',
+        '#include "trace/simple.h"',
         '',
         'TraceEvent trace_list[] = {')
 
@@ -26,30 +33,69 @@ def c(events):
             name = e.name,
             )
 
-    out('};')
-
-def h(events):
-    out('#include "trace/simple.h"',
+    out('};',
         '')
 
-    for num, e in enumerate(events):
-        if len(e.args):
-            argstr = e.args.names()
-            arg_prefix = ', (uint64_t)(uintptr_t)'
-            cast_args = arg_prefix + arg_prefix.join(argstr)
-            simple_args = (str(num) + cast_args)
-        else:
-            simple_args = str(num)
+    for num, event in enumerate(events):
+        sizes = []
+        for type_, name in event.args:
+            if is_string(type_):
+                sizes.append("4 + (" + name + " ? strlen(" + name + ") : 0)")
+            else:
+                sizes.append("8")
+        sizestr = " + ".join(sizes)
+        if len(event.args) == 0:
+            sizestr = '0'
 
-        out('static inline void trace_%(name)s(%(args)s)',
+        out('void trace_%(name)s(%(args)s)',
             '{',
-            '    trace%(argc)d(%(trace_args)s);',
-            '}',
-            name = e.name,
-            args = e.args,
-            argc = len(e.args),
-            trace_args = simple_args,
+            '    TraceBufferRecord rec;',
+            '',
+            '    if (!trace_list[%(event_id)s].state) {',
+            '        return;',
+            '    }',
+            '',
+            '    if (trace_record_start(&rec, %(event_id)s, %(size_str)s)) {',
+            '        return; /* Trace Buffer Full, Event Dropped ! */',
+            '    }',
+            name = event.name,
+            args = event.args,
+            event_id = num,
+            size_str = sizestr,
             )
 
+        if len(event.args) > 0:
+            for type_, name in event.args:
+                # string
+                if is_string(type_):
+                    out('    trace_record_write_str(&rec, %(name)s);',
+                        name = name,
+                       )
+                # pointer var (not string)
+                elif type_.endswith('*'):
+                    out('    trace_record_write_u64(&rec, (uint64_t)(uint64_t *)%(name)s);',
+                        name = name,
+                       )
+                # primitive data type
+                else:
+                    out('    trace_record_write_u64(&rec, (uint64_t)%(name)s);',
+                       name = name,
+                       )
+
+        out('    trace_record_finish(&rec);',
+            '}',
+            '')
+
+
+def h(events):
+    out('#include "trace/simple.h"',
+        '')
+
+    for event in events:
+        out('void trace_%(name)s(%(args)s);',
+            name = event.name,
+            args = event.args,
+            )
+    out('')
     out('#define NR_TRACE_EVENTS %d' % len(events))
     out('extern TraceEvent trace_list[NR_TRACE_EVENTS];')
diff --git a/trace/simple.c b/trace/simple.c
index b64bcf4..988a1e8 100644
--- a/trace/simple.c
+++ b/trace/simple.c
@@ -27,7 +27,7 @@
 #define HEADER_MAGIC 0xf2b177cb0aa429b4ULL
 
 /** Trace file version number, bump if format changes */
-#define HEADER_VERSION 0
+#define HEADER_VERSION 2
 
 /** Records were dropped event ID */
 #define DROPPED_EVENT_ID (~(uint64_t)0 - 1)
@@ -35,23 +35,6 @@
 /** Trace record is valid */
 #define TRACE_RECORD_VALID ((uint64_t)1 << 63)
 
-/** Trace buffer entry */
-typedef struct {
-    uint64_t event;
-    uint64_t timestamp_ns;
-    uint64_t x1;
-    uint64_t x2;
-    uint64_t x3;
-    uint64_t x4;
-    uint64_t x5;
-    uint64_t x6;
-} TraceRecord;
-
-enum {
-    TRACE_BUF_LEN = 4096,
-    TRACE_BUF_FLUSH_THRESHOLD = TRACE_BUF_LEN / 4,
-};
-
 /*
  * Trace records are written out by a dedicated thread.  The thread waits for
  * records to become available, writes them out, and then waits again.
@@ -62,11 +45,39 @@ static GCond *trace_empty_cond;
 static bool trace_available;
 static bool trace_writeout_enabled;
 
-static TraceRecord trace_buf[TRACE_BUF_LEN];
+enum {
+    TRACE_BUF_LEN = 4096 * 64,
+    TRACE_BUF_FLUSH_THRESHOLD = TRACE_BUF_LEN / 4,
+};
+
+uint8_t trace_buf[TRACE_BUF_LEN];
 static unsigned int trace_idx;
+static unsigned int writeout_idx;
+static uint64_t dropped_events;
 static FILE *trace_fp;
 static char *trace_file_name = NULL;
 
+/* * Trace buffer entry */
+typedef struct {
+    uint64_t event; /*   TraceEventID */
+    uint64_t timestamp_ns;
+    uint32_t length;   /*    in bytes */
+    uint32_t reserved; /*    unused */
+    uint8_t arguments[];
+} TraceRecord;
+
+typedef struct {
+    uint64_t header_event_id; /* HEADER_EVENT_ID */
+    uint64_t header_magic;    /* HEADER_MAGIC    */
+    uint64_t header_version;  /* HEADER_VERSION  */
+} TraceRecordHeader;
+
+
+int trace_alloc_record(TraceBufferRecord *rec, TraceEventID event, uint32_t datasize);
+static void read_from_buffer(unsigned int idx, void *dataptr, size_t size);
+static unsigned int write_to_buffer(unsigned int idx, void *dataptr, size_t size);
+void trace_mark_record_complete(TraceBufferRecord *rec);
+
 /**
  * Read a trace record from the trace buffer
  *
@@ -75,16 +86,27 @@ static char *trace_file_name = NULL;
  *
  * Returns false if the record is not valid.
  */
-static bool get_trace_record(unsigned int idx, TraceRecord *record)
+static bool get_trace_record(unsigned int idx, TraceRecord **recordptr)
 {
-    if (!(trace_buf[idx].event & TRACE_RECORD_VALID)) {
+    uint8_t rec_hdr[sizeof(TraceRecord)];
+    uint64_t event_flag = 0;
+    TraceRecord *record = (TraceRecord *) rec_hdr;
+    /* read the event flag to see if its a valid record */
+    read_from_buffer(idx, rec_hdr, sizeof(event_flag));
+
+    if (!(record->event & TRACE_RECORD_VALID)) {
         return false;
     }
 
     __sync_synchronize(); /* read memory barrier before accessing record */
-
-    *record = trace_buf[idx];
-    record->event &= ~TRACE_RECORD_VALID;
+    /* read the record header to know record length */
+    read_from_buffer(idx, rec_hdr, sizeof(TraceRecord));
+    *recordptr = g_malloc(record->length);
+    /* make a copy of record to avoid being overwritten */
+    read_from_buffer(idx, (uint8_t *)*recordptr, record->length);
+    (*recordptr)->event &= ~TRACE_RECORD_VALID;
+    /* reset record event validity flag on global trace buffer */
+    write_to_buffer(idx, &event_flag, sizeof(event_flag));
     return true;
 }
 
@@ -120,29 +142,30 @@ static void wait_for_trace_records_available(void)
 
 static gpointer writeout_thread(gpointer opaque)
 {
-    TraceRecord record;
-    unsigned int writeout_idx = 0;
-    unsigned int num_available, idx;
+    TraceRecord *recordptr, *dropped_ptr;
+    uint8_t dropped_rec[sizeof(TraceRecord) + sizeof(dropped_events)];
+    dropped_ptr = (TraceRecord *)dropped_rec;
+    unsigned int idx = 0;
     size_t unused __attribute__ ((unused));
 
     for (;;) {
         wait_for_trace_records_available();
 
-        num_available = trace_idx - writeout_idx;
-        if (num_available > TRACE_BUF_LEN) {
-            record = (TraceRecord){
-                .event = DROPPED_EVENT_ID,
-                .x1 = num_available,
-            };
-            unused = fwrite(&record, sizeof(record), 1, trace_fp);
-            writeout_idx += num_available;
+        if (dropped_events) {
+            dropped_ptr->event = DROPPED_EVENT_ID,
+            dropped_ptr->timestamp_ns = get_clock();
+            dropped_ptr->length = sizeof(TraceRecord) + sizeof(dropped_events),
+            dropped_ptr->reserved = 0;
+            memcpy(dropped_ptr->arguments, &dropped_events, sizeof(dropped_events));
+            dropped_events = 0;
+            unused = fwrite(dropped_ptr, dropped_ptr->length, 1, trace_fp);
         }
 
-        idx = writeout_idx % TRACE_BUF_LEN;
-        while (get_trace_record(idx, &record)) {
-            trace_buf[idx].event = 0; /* clear valid bit */
-            unused = fwrite(&record, sizeof(record), 1, trace_fp);
-            idx = ++writeout_idx % TRACE_BUF_LEN;
+        while (get_trace_record(idx, &recordptr)) {
+            unused = fwrite(recordptr, recordptr->length, 1, trace_fp);
+            writeout_idx += recordptr->length;
+            g_free(recordptr);
+            idx = writeout_idx % TRACE_BUF_LEN;
         }
 
         fflush(trace_fp);
@@ -150,73 +173,102 @@ static gpointer writeout_thread(gpointer opaque)
     return NULL;
 }
 
-static void trace(TraceEventID event, uint64_t x1, uint64_t x2, uint64_t x3,
-                  uint64_t x4, uint64_t x5, uint64_t x6)
+int trace_record_start(TraceBufferRecord *rec, TraceEventID id, size_t arglen)
 {
-    unsigned int idx;
-    uint64_t timestamp;
-
-    if (!trace_list[event].state) {
-        return;
-    }
-
-    timestamp = get_clock();
-#if GLIB_CHECK_VERSION(2, 30, 0)
-    idx = g_atomic_int_add((gint *)&trace_idx, 1) % TRACE_BUF_LEN;
-#else
-    idx = g_atomic_int_exchange_and_add((gint *)&trace_idx, 1) % TRACE_BUF_LEN;
-#endif
-    trace_buf[idx] = (TraceRecord){
-        .event = event,
-        .timestamp_ns = timestamp,
-        .x1 = x1,
-        .x2 = x2,
-        .x3 = x3,
-        .x4 = x4,
-        .x5 = x5,
-        .x6 = x6,
-    };
-    __sync_synchronize(); /* write barrier before marking as valid */
-    trace_buf[idx].event |= TRACE_RECORD_VALID;
-
-    if ((idx + 1) % TRACE_BUF_FLUSH_THRESHOLD == 0) {
-        flush_trace_file(false);
-    }
+    return trace_alloc_record(rec, id, arglen); /* return 0 on success */
 }
 
-void trace0(TraceEventID event)
+void trace_record_write_u64(TraceBufferRecord *rec, uint64_t val)
 {
-    trace(event, 0, 0, 0, 0, 0, 0);
+    rec->rec_off = write_to_buffer(rec->rec_off, (uint8_t *)&val, sizeof(uint64_t));
 }
 
-void trace1(TraceEventID event, uint64_t x1)
+void trace_record_write_str(TraceBufferRecord *rec, const char *s)
 {
-    trace(event, x1, 0, 0, 0, 0, 0);
+    /* Write string length first */
+    uint32_t slen = (s == NULL ? 0 : strlen(s));
+    rec->rec_off = write_to_buffer(rec->rec_off, (uint8_t *)&slen, sizeof(slen));
+    /* Write actual string now */
+    rec->rec_off = write_to_buffer(rec->rec_off, (uint8_t *)s, slen);
 }
 
-void trace2(TraceEventID event, uint64_t x1, uint64_t x2)
+void trace_record_finish(TraceBufferRecord *rec)
 {
-    trace(event, x1, x2, 0, 0, 0, 0);
+    trace_mark_record_complete(rec);
 }
 
-void trace3(TraceEventID event, uint64_t x1, uint64_t x2, uint64_t x3)
+int trace_alloc_record(TraceBufferRecord *rec, TraceEventID event, uint32_t datasize)
 {
-    trace(event, x1, x2, x3, 0, 0, 0);
+    unsigned int idx, rec_off, old_idx, new_idx;
+    uint32_t rec_len = sizeof(TraceRecord) + datasize;
+    uint64_t timestamp_ns = get_clock();
+
+    while (1) {
+        old_idx = trace_idx;
+        smp_rmb();
+        new_idx = old_idx + rec_len;
+
+        if (new_idx - writeout_idx > TRACE_BUF_LEN) {
+            dropped_events++; /* Trace Buffer Full, Event dropped ! */
+            return - ENOSPC;
+        }
+
+        if g_atomic_int_compare_and_exchange((gint *)&trace_idx, old_idx, new_idx)
+            break;
+    }
+
+    idx = old_idx % TRACE_BUF_LEN;
+    /*  To check later if threshold crossed */
+    rec->next_tbuf_idx = trace_idx % TRACE_BUF_LEN;
+
+    rec_off = idx;
+    rec_off = write_to_buffer(rec_off, (uint8_t*)&event, sizeof(event));
+    rec_off = write_to_buffer(rec_off, (uint8_t*)&timestamp_ns, sizeof(timestamp_ns));
+    rec_off = write_to_buffer(rec_off, (uint8_t*)&rec_len, sizeof(rec_len));
+
+    rec->tbuf_idx = idx;
+    rec->rec_off  = (idx + sizeof(TraceRecord)) % TRACE_BUF_LEN;
+    return 0;
 }
 
-void trace4(TraceEventID event, uint64_t x1, uint64_t x2, uint64_t x3, uint64_t x4)
+static void read_from_buffer(unsigned int idx, void *dataptr, size_t size)
 {
-    trace(event, x1, x2, x3, x4, 0, 0);
+    uint8_t *data_ptr = dataptr;
+    uint32_t x = 0;
+    while (x < size) {
+        if (idx >= TRACE_BUF_LEN) {
+            idx = idx % TRACE_BUF_LEN;
+        }
+        data_ptr[x++] = trace_buf[idx++];
+    }
 }
 
-void trace5(TraceEventID event, uint64_t x1, uint64_t x2, uint64_t x3, uint64_t x4, uint64_t x5)
+static unsigned int write_to_buffer(unsigned int idx, void *dataptr, size_t size)
 {
-    trace(event, x1, x2, x3, x4, x5, 0);
+    uint8_t *data_ptr = dataptr;
+    uint32_t x = 0;
+    while (x < size) {
+        if (idx >= TRACE_BUF_LEN) {
+            idx = idx % TRACE_BUF_LEN;
+        }
+        trace_buf[idx++] = data_ptr[x++];
+    }
+    return idx; /* most callers wants to know where to write next */
 }
 
-void trace6(TraceEventID event, uint64_t x1, uint64_t x2, uint64_t x3, uint64_t x4, uint64_t x5, uint64_t x6)
+void trace_mark_record_complete(TraceBufferRecord *rec)
 {
-    trace(event, x1, x2, x3, x4, x5, x6);
+    uint8_t temp_rec[sizeof(TraceRecord)];
+    TraceRecord *record = (TraceRecord *) temp_rec;
+    read_from_buffer(rec->tbuf_idx, temp_rec, sizeof(TraceRecord));
+    __sync_synchronize(); /* write barrier before marking as valid */
+    record->event |= TRACE_RECORD_VALID;
+    write_to_buffer(rec->tbuf_idx, temp_rec, sizeof(TraceRecord));
+
+    if (rec->next_tbuf_idx > TRACE_BUF_FLUSH_THRESHOLD &&
+        rec->tbuf_idx <= TRACE_BUF_FLUSH_THRESHOLD) {
+        flush_trace_file(false);
+    }
 }
 
 void st_set_trace_file_enabled(bool enable)
@@ -231,10 +283,11 @@ void st_set_trace_file_enabled(bool enable)
     flush_trace_file(true);
 
     if (enable) {
-        static const TraceRecord header = {
-            .event = HEADER_EVENT_ID,
-            .timestamp_ns = HEADER_MAGIC,
-            .x1 = HEADER_VERSION,
+        static const TraceRecordHeader header = {
+            .header_event_id = HEADER_EVENT_ID,
+            .header_magic = HEADER_MAGIC,
+            /* Older log readers will check for version at next location */
+            .header_version = HEADER_VERSION,
         };
 
         trace_fp = fopen(trace_file_name, "wb");
diff --git a/trace/simple.h b/trace/simple.h
index 6b5358c..6755e99 100644
--- a/trace/simple.h
+++ b/trace/simple.h
@@ -22,16 +22,39 @@ typedef struct {
     bool state;
 } TraceEvent;
 
-void trace0(TraceEventID event);
-void trace1(TraceEventID event, uint64_t x1);
-void trace2(TraceEventID event, uint64_t x1, uint64_t x2);
-void trace3(TraceEventID event, uint64_t x1, uint64_t x2, uint64_t x3);
-void trace4(TraceEventID event, uint64_t x1, uint64_t x2, uint64_t x3, uint64_t x4);
-void trace5(TraceEventID event, uint64_t x1, uint64_t x2, uint64_t x3, uint64_t x4, uint64_t x5);
-void trace6(TraceEventID event, uint64_t x1, uint64_t x2, uint64_t x3, uint64_t x4, uint64_t x5, uint64_t x6);
 void st_print_trace_file_status(FILE *stream, fprintf_function stream_printf);
 void st_set_trace_file_enabled(bool enable);
 bool st_set_trace_file(const char *file);
 void st_flush_trace_buffer(void);
 
+typedef struct {
+    unsigned int tbuf_idx;
+    unsigned int next_tbuf_idx;
+    unsigned int rec_off;
+} TraceBufferRecord;
+
+/**
+ * Initialize a trace record and claim space for it in the buffer
+ *
+ * @arglen  number of bytes required for arguments
+ */
+int trace_record_start(TraceBufferRecord *rec, TraceEventID id, size_t arglen);
+
+/**
+ * Append a 64-bit argument to a trace record
+ */
+void trace_record_write_u64(TraceBufferRecord *rec, uint64_t val);
+
+/**
+ * Append a string argument to a trace record
+ */
+void trace_record_write_str(TraceBufferRecord *rec, const char *s);
+
+/**
+ * Mark a trace record completed
+ *
+ * Don't append any more arguments to the trace record after calling this.
+ */
+void trace_record_finish(TraceBufferRecord *rec);
+
 #endif /* TRACE_SIMPLE_H */
-- 
1.7.10.2

[Qemu-devel] [PATCH v4 3/3] Update simpletrace.py for new log format

[Harsh Prateek Bora]

Support new tracelog format for multiple arguments and strings.

Signed-off-by: Harsh Prateek Bora <harsh@linux.vnet.ibm.com>
---
 scripts/simpletrace.py |  116 +++++++++++++++++++++++++++++++-----------------
 1 file changed, 75 insertions(+), 41 deletions(-)

diff --git a/scripts/simpletrace.py b/scripts/simpletrace.py
index f55e5e6..9b4419f 100755
--- a/scripts/simpletrace.py
+++ b/scripts/simpletrace.py
@@ -12,53 +12,69 @@
 import struct
 import re
 import inspect
+from tracetool import _read_events, Event
+from tracetool.backend.simple import is_string
 
 header_event_id = 0xffffffffffffffff
 header_magic    = 0xf2b177cb0aa429b4
-header_version  = 0
 dropped_event_id = 0xfffffffffffffffe
 
-trace_fmt = '=QQQQQQQQ'
-trace_len = struct.calcsize(trace_fmt)
-event_re  = re.compile(r'(disable\s+)?([a-zA-Z0-9_]+)\(([^)]*)\).*')
+log_header_fmt = '=QQQ'
+rec_header_fmt = '=QQII'
 
-def parse_events(fobj):
-    """Parse a trace-events file into {event_num: (name, arg1, ...)}."""
-
-    def get_argnames(args):
-        """Extract argument names from a parameter list."""
-        return tuple(arg.split()[-1].lstrip('*') for arg in args.split(','))
-
-    events = {dropped_event_id: ('dropped', 'count')}
-    event_num = 0
-    for line in fobj:
-        m = event_re.match(line.strip())
-        if m is None:
-            continue
-
-        disable, name, args = m.groups()
-        events[event_num] = (name,) + get_argnames(args)
-        event_num += 1
-    return events
+def read_header(fobj, hfmt):
+    '''Read a trace record header'''
+    hlen = struct.calcsize(hfmt)
+    hdr = fobj.read(hlen)
+    if len(hdr) != hlen:
+        return None
+    return struct.unpack(hfmt, hdr)
 
-def read_record(fobj):
+def get_record(edict, rechdr, fobj):
     """Deserialize a trace record from a file into a tuple (event_num, timestamp, arg1, ..., arg6)."""
-    s = fobj.read(trace_len)
-    if len(s) != trace_len:
+    if rechdr is None:
         return None
-    return struct.unpack(trace_fmt, s)
+    rec = (rechdr[0], rechdr[1])
+    if rechdr[0] != dropped_event_id:
+        event_id = rechdr[0]
+        event = edict[event_id]
+        for type, name in event.args:
+            if is_string(type):
+                l = fobj.read(4)
+                (len,) = struct.unpack('=L', l)
+                s = fobj.read(len)
+                rec = rec + (s,)
+            else:
+                (value,) = struct.unpack('=Q', fobj.read(8))
+                rec = rec + (value,)
+    else:
+        (value,) = struct.unpack('=Q', fobj.read(8))
+        rec = rec + (value,)
+    return rec
+
+
+def read_record(edict, fobj):
+    """Deserialize a trace record from a file into a tuple (event_num, timestamp, arg1, ..., arg6)."""
+    rechdr = read_header(fobj, rec_header_fmt)
+    return get_record(edict, rechdr, fobj) # return tuple of record elements
 
-def read_trace_file(fobj):
+def read_trace_file(edict, fobj):
     """Deserialize trace records from a file, yielding record tuples (event_num, timestamp, arg1, ..., arg6)."""
-    header = read_record(fobj)
+    header = read_header(fobj, log_header_fmt)
     if header is None or \
        header[0] != header_event_id or \
-       header[1] != header_magic or \
-       header[2] != header_version:
-        raise ValueError('not a trace file or incompatible version')
+       header[1] != header_magic:
+        raise ValueError('Not a valid trace file!')
+    if header[2] != 0 and \
+       header[2] != 2:
+        raise ValueError('Unknown version of tracelog format!')
+
+    log_version = header[2]
+    if log_version == 0:
+        raise ValueError('Older log format, not supported with this Qemu release!')
 
     while True:
-        rec = read_record(fobj)
+        rec = read_record(edict, fobj)
         if rec is None:
             break
 
@@ -89,16 +105,29 @@ class Analyzer(object):
 def process(events, log, analyzer):
     """Invoke an analyzer on each event in a log."""
     if isinstance(events, str):
-        events = parse_events(open(events, 'r'))
+        events = _read_events(open(events, 'r'))
     if isinstance(log, str):
         log = open(log, 'rb')
 
+    enabled_events = []
+    dropped_event = Event.build("Dropped_Event(uint64_t num_events_dropped)")
+    edict = {dropped_event_id: dropped_event}
+
+    for e in events:
+        if 'disable' not in e.properties:
+            enabled_events.append(e)
+    for num, event in enumerate(enabled_events):
+        edict[num] = event
+
     def build_fn(analyzer, event):
-        fn = getattr(analyzer, event[0], None)
+        if isinstance(event, str):
+            return analyzer.catchall
+
+        fn = getattr(analyzer, event.name, None)
         if fn is None:
             return analyzer.catchall
 
-        event_argcount = len(event) - 1
+        event_argcount = len(event.args)
         fn_argcount = len(inspect.getargspec(fn)[0]) - 1
         if fn_argcount == event_argcount + 1:
             # Include timestamp as first argument
@@ -109,9 +138,9 @@ def process(events, log, analyzer):
 
     analyzer.begin()
     fn_cache = {}
-    for rec in read_trace_file(log):
+    for rec in read_trace_file(edict, log):
         event_num = rec[0]
-        event = events[event_num]
+        event = edict[event_num]
         if event_num not in fn_cache:
             fn_cache[event_num] = build_fn(analyzer, event)
         fn_cache[event_num](event, rec)
@@ -128,7 +157,7 @@ def run(analyzer):
         sys.stderr.write('usage: %s <trace-events> <trace-file>\n' % sys.argv[0])
         sys.exit(1)
 
-    events = parse_events(open(sys.argv[1], 'r'))
+    events = _read_events(open(sys.argv[1], 'r'))
     process(events, sys.argv[2], analyzer)
 
 if __name__ == '__main__':
@@ -137,15 +166,20 @@ if __name__ == '__main__':
             self.last_timestamp = None
 
         def catchall(self, event, rec):
+            i = 1
             timestamp = rec[1]
             if self.last_timestamp is None:
                 self.last_timestamp = timestamp
             delta_ns = timestamp - self.last_timestamp
             self.last_timestamp = timestamp
 
-            fields = [event[0], '%0.3f' % (delta_ns / 1000.0)]
-            for i in xrange(1, len(event)):
-                fields.append('%s=0x%x' % (event[i], rec[i + 1]))
+            fields = [event.name, '%0.3f' % (delta_ns / 1000.0)]
+            for type, name in event.args:
+                if is_string(type):
+                    fields.append('%s=%s' % (name, rec[i + 1]))
+                else:
+                    fields.append('%s=0x%x' % (name, rec[i + 1]))
+                i += 1
             print ' '.join(fields)
 
     run(Formatter())
-- 
1.7.10.2

Re: [Qemu-devel] [PATCH v4 2/3] Simpletrace v2: Support multiple arguments, strings.

[Stefan Hajnoczi]

On Tue, Jun 12, 2012 at 4:16 PM, Harsh Prateek Bora
<harsh@linux.vnet.ibm.com> wrote:
> +int trace_alloc_record(TraceBufferRecord *rec, TraceEventID event, uint32_t datasize);

This should be static.

> +static void read_from_buffer(unsigned int idx, void *dataptr, size_t size);
> +static unsigned int write_to_buffer(unsigned int idx, void *dataptr, size_t size);
> +void trace_mark_record_complete(TraceBufferRecord *rec);

This should be static.  Can we get rid of these internal names
completely?  trace_record_finish() could hold the code instead of
calling into an internal function.  Same for trace_record_start()
above.

> +
>  /**
>  * Read a trace record from the trace buffer
>  *
> @@ -75,16 +86,27 @@ static char *trace_file_name = NULL;
>  *
>  * Returns false if the record is not valid.
>  */
> -static bool get_trace_record(unsigned int idx, TraceRecord *record)
> +static bool get_trace_record(unsigned int idx, TraceRecord **recordptr)
>  {
> -    if (!(trace_buf[idx].event & TRACE_RECORD_VALID)) {
> +    uint8_t rec_hdr[sizeof(TraceRecord)];
> +    uint64_t event_flag = 0;
> +    TraceRecord *record = (TraceRecord *) rec_hdr;
> +    /* read the event flag to see if its a valid record */
> +    read_from_buffer(idx, rec_hdr, sizeof(event_flag));
> +
> +    if (!(record->event & TRACE_RECORD_VALID)) {
>         return false;
>     }
>
>     __sync_synchronize(); /* read memory barrier before accessing record */
> -
> -    *record = trace_buf[idx];
> -    record->event &= ~TRACE_RECORD_VALID;
> +    /* read the record header to know record length */
> +    read_from_buffer(idx, rec_hdr, sizeof(TraceRecord));
> +    *recordptr = g_malloc(record->length);
> +    /* make a copy of record to avoid being overwritten */
> +    read_from_buffer(idx, (uint8_t *)*recordptr, record->length);
> +    (*recordptr)->event &= ~TRACE_RECORD_VALID;

Here I think we need another barrier - we don't want to read data
after clearing the valid flag (otherwise we might read data for a new
event, not the event we just read!).  This is being superparanoid but
let's be careful for starters and if someone can explain why it's safe
to eliminate the barrier, then we can remove it.

> +    /* reset record event validity flag on global trace buffer */
> +    write_to_buffer(idx, &event_flag, sizeof(event_flag));
>     return true;
>  }
>
> @@ -120,29 +142,30 @@ static void wait_for_trace_records_available(void)
>
>  static gpointer writeout_thread(gpointer opaque)
>  {
> -    TraceRecord record;
> -    unsigned int writeout_idx = 0;
> -    unsigned int num_available, idx;
> +    TraceRecord *recordptr, *dropped_ptr;
> +    uint8_t dropped_rec[sizeof(TraceRecord) + sizeof(dropped_events)];
> +    dropped_ptr = (TraceRecord *)dropped_rec;

Since you need to resend, please use a union instead of a uint8_t
array and a pointer.  This tells the compiler your intention and
guarantees that the memory is aligned properly on all host platforms.

union {
    TraceRecord rec;
    uint8_t bytes[sizeof(TraceRecord) + sizeof(uint64_t)];
} dropped;

> +    unsigned int idx = 0;
>     size_t unused __attribute__ ((unused));
>
>     for (;;) {
>         wait_for_trace_records_available();
>
> -        num_available = trace_idx - writeout_idx;
> -        if (num_available > TRACE_BUF_LEN) {
> -            record = (TraceRecord){
> -                .event = DROPPED_EVENT_ID,
> -                .x1 = num_available,
> -            };
> -            unused = fwrite(&record, sizeof(record), 1, trace_fp);
> -            writeout_idx += num_available;
> +        if (dropped_events) {
> +            dropped_ptr->event = DROPPED_EVENT_ID,
> +            dropped_ptr->timestamp_ns = get_clock();
> +            dropped_ptr->length = sizeof(TraceRecord) + sizeof(dropped_events),
> +            dropped_ptr->reserved = 0;
> +            memcpy(dropped_ptr->arguments, &dropped_events, sizeof(dropped_events));
> +            dropped_events = 0;

To avoid losing counts we need cmpxchg &dropped_events, 0.  Remember
one or more threads might be writing to dropped_events while this code
executes!  Since this value is a count, not a boolean, let's ensure it
is accurate.

> +            unused = fwrite(dropped_ptr, dropped_ptr->length, 1, trace_fp);
>         }
>
> -        idx = writeout_idx % TRACE_BUF_LEN;
> -        while (get_trace_record(idx, &record)) {
> -            trace_buf[idx].event = 0; /* clear valid bit */
> -            unused = fwrite(&record, sizeof(record), 1, trace_fp);
> -            idx = ++writeout_idx % TRACE_BUF_LEN;
> +        while (get_trace_record(idx, &recordptr)) {
> +            unused = fwrite(recordptr, recordptr->length, 1, trace_fp);
> +            writeout_idx += recordptr->length;
> +            g_free(recordptr);
> +            idx = writeout_idx % TRACE_BUF_LEN;
>         }

This has become quite nice!

> -void trace3(TraceEventID event, uint64_t x1, uint64_t x2, uint64_t x3)
> +int trace_alloc_record(TraceBufferRecord *rec, TraceEventID event, uint32_t datasize)
>  {
> -    trace(event, x1, x2, x3, 0, 0, 0);
> +    unsigned int idx, rec_off, old_idx, new_idx;
> +    uint32_t rec_len = sizeof(TraceRecord) + datasize;
> +    uint64_t timestamp_ns = get_clock();
> +
> +    while (1) {
> +        old_idx = trace_idx;
> +        smp_rmb();
> +        new_idx = old_idx + rec_len;
> +
> +        if (new_idx - writeout_idx > TRACE_BUF_LEN) {
> +            dropped_events++; /* Trace Buffer Full, Event dropped ! */

See comment above about making dropped_events atomic.

> +            return - ENOSPC;

Just a syntax nitpick but it's very unusual to put a space after the
negative, it looks like you were trying to use the binary minus
operator.  The usual way is "-ENOSPC".

> +        }
> +
> +        if g_atomic_int_compare_and_exchange((gint *)&trace_idx, old_idx, new_idx)
> +            break;

Please run scripts/checkpatch.pl.  QEMU always uses {} for bodies of
if/while statements.

> +    }
> +
> +    idx = old_idx % TRACE_BUF_LEN;
> +    /*  To check later if threshold crossed */
> +    rec->next_tbuf_idx = trace_idx % TRACE_BUF_LEN;

Careful, this looks wrong since trace_idx can change while this
function is executing.  Did you mean new_idx?

> -void trace6(TraceEventID event, uint64_t x1, uint64_t x2, uint64_t x3, uint64_t x4, uint64_t x5, uint64_t x6)
> +void trace_mark_record_complete(TraceBufferRecord *rec)
>  {
> -    trace(event, x1, x2, x3, x4, x5, x6);
> +    uint8_t temp_rec[sizeof(TraceRecord)];
> +    TraceRecord *record = (TraceRecord *) temp_rec;
> +    read_from_buffer(rec->tbuf_idx, temp_rec, sizeof(TraceRecord));
> +    __sync_synchronize(); /* write barrier before marking as valid */
> +    record->event |= TRACE_RECORD_VALID;
> +    write_to_buffer(rec->tbuf_idx, temp_rec, sizeof(TraceRecord));
> +
> +    if (rec->next_tbuf_idx > TRACE_BUF_FLUSH_THRESHOLD &&
> +        rec->tbuf_idx <= TRACE_BUF_FLUSH_THRESHOLD) {
> +        flush_trace_file(false);
> +    }

There is an important difference between this new code and the old
code.  Previously we did modulo TRACE_BUF_FLUSH_THRESHOLD, this cause
a flush *every* TRACE_BUF_FLUSH_THRESHOLD entries.  The new code only
flushes once: when we cross the TRACE_BUF_FLUSH_THRESHOLD byte
boundary.

This new flush condition is no better than flushing when we wrap the
buffer.  In other words, the buffer will be full when we flush!  We
need to avoid this so that the chance of events getting dropped is
low.

The idea of checking when we cross the threshold is good, but it needs
to apply *every* TRACE_BUF_FLUSH_THRESHOLD bytes, not just once.

Stefan